[From Bill Powers (931221.1130 MST)]
Osmo Eerola (931221.0810 GMT) --
Why do not you use just the term CT? Everything would me much
easier to stupid control engineers. The brand new insight of
controlling perceptions could well be included in the framework
of the (traditional) CT. But I guess that you need a new term
to make a high profile in behavioral sciences.
The reason is to distinguish our use of control theory for
explaining the behavior of organisms from the uses that are found
in the literature of engineering control. A psychologist who
tried to look up control theory textbooks would find little help
there -- psychologists have been doing that for 40 years,
regularly being misled by what they find. At least when they see
_perceptual_ control theory, they know there is something funny
about it, and they may pay closer attention to the details.
... the term diff = ref-P is used for driving (controlling) the
process. Therefore the control of perception (P) sounds so
strange to control engineers, but as said many times, one can
say so.
I distinguish between driving a process and controlling it. You
can drive a process open-loop, but if there are any disturbances
or changes in parameters, you can't control it open-loop. To me,
controlling means causing a variable to approach a preselected
state and to stay in that state even when it's influenced by
independent disturbances. It means being able to change the
desired state and have the variable also change to the new state,
again even when disturbances are acting. If there are such
disturbances, then to maintain the SAME state of a process, you
must VARY the drive to the process.
I think of controlling as something done by the whole closed
loop, not by any one part of it. What part could you omit? Even
in so-called open loop control processes, the loop is really
closed. It is closed because somebody can see the state of the
controlled process, and can alter the drive mechanism in the
forward part of the loop until the controlled process is
perceived in the desired state. Setting up an open-loop process
so it will continue working properly under a variety of drive
conditions requires a closed-loop process: observing the result,
adjusting the system, observing, adjusting, and so on. All that
the open-loop system gains for you is the ability to leave it
running for a while without paying attention to it. It is
unsuitable for any situation where unexpected disturbances can
arise. In human behavior, that means almost every situation.
So basically you are controlling with the computing element
(neurons), which derives information from the sensory
receptors, the muscular effectors (process, plant). You say,
however, you control perception. Okay, the whole system as
described above, still seems to me very conventional control
approach - and there is NOTHING WRONG in it, just fine!
I'm glad to hear "just fine" from a real control engineer. In
spite of what I think I know about control, it's still
reassuring.
The unconventional part is to focus on perception as the thing
being controlled, instead of the external process that's being
perceived. Beside the reasons I've already given, there is
another that we haven't talked about. Perceptions are not
necessarily in one-to-one correspondence with observable physical
variables, and what you want to control might not be a tangible
physical thing or process.
Consider controlling the distance between two objects by moving
one or both of them. The position of each object can be observed
against the background, but how do you observe the distance
between them? Where the distance is, there is no object. There is
no way to act on the distance; all you can act upon is an object.
So imagine a control system with a more complex input function.
This input function can perceive the positions of two objects and
report the positions as two signals. The signals are each
proportional in magnitude to a reading on a position scale. With
that information, you could control the position of either
object, but not the distance between them.
To control the distance between them, you would have to add
another layer of perception, a device that would subtract one
signal from the other to produce a new signal indicating the
difference in positions, independently of the absolute positions
on the scale. Now this distance-perception could be controlled by
comparing it with a reference signal indicating the desired
distance between the positions, and turning the error into a
force that moves one or both objects. We can now control a
variable that has no physical existence in the environment -- a
variable that corresponds to no mass and no energy, an idea.
You'll notice that we aren't talking about a simple G(s) or H(s)
any more -- we're talking about a system with multiple inputs and
outputs that can vary independently of each other.
This is just a very simple example. It's possible to come up with
much more complex ones, which can really be understood best in
terms of controlling perceptions, not things or processes.
I would guess that engineers occasionally have the problem of
controlling the energy efficiency of a process. Energy efficiency
isn't a physical thing; it's a calculation. Efficiency is a
function of many process variables, so it would be possible to
construct a perceptual function that would compute and output a
signal corresponding to energy efficiency. The desired energy
efficiency might be some specific number less than the maximum
achievable, for example because pollution also has to be
considered, or cost, or the avoidance of excessive temperatures.
Given a perceptual signal that indicates the present level of
efficiency, it could be compared with a reference signal
indicating the desired level, and the error could then be
converted into the appropriate directions of change of the
individual process variables, each direction chosen to produce a
negative feedback effect on the perception of efficiency, with
suitable filtering for stability. Now a system with multiple
outputs is being used to control a perceptual signal that is a
function of many individual inputs. It is not controlling any one
of those physical variables, but a derived variable that is a
function of them.
Something similar to this is done in modern automobile control
computers, which alter several operational variables as a way of
reducing exhaust emissions while retaining good gas mileage, a
compromise that requires controlling a variable that is a
function of several other variables -- a computed variable, a
perception.
A different way of looking, not a new science.
Right. But it is a useful way of looking, and even in control
engineering it might help to organize the design of complex
control systems. I have seen many designs in which the engineer's
intention was clear, but in which that intention failed to become
an intention of the control system itself. If the control system
could be equipped to perceive the same variable that the engineer
is trying to control, then perhaps the control system itself
could achieve the desired end automatically, instead of just
being made to act as desired by clever -- and often obscure --
tricks of design.
I remember a design in which twin helicopters were to be used
under common control to lift heavy loads. But there was nothing
in the control system that sensed whether the rotors would hit
each other. Instead, the engineer adjusted certain constants,
like the lift angles, to make sure that such collisions
(probably) wouldn't occur. It would have been better for the
control system to be given information such as the angles of the
supporting cables and the length of free cable, so it could
perceive the distance between the helicopters and control it to
maintain the distance greater than the rotor diameters. Thinking
about control systems as having perceptions can lead us to think
about giving them useful perceptions that they can control for
themselves. In many designs I have seen, variables that should be
under control are controlled only in the design engineer's head.
Nothing new, as you say. But perhaps this way of looking at
things might prove valuable even in engineering.
ยทยทยท
--------------------------
Old hat S-R system. Not totally wrong: when I see an angry dog
(stimulus) I react (response) by doing something.
When you get used to thinking in PCT terms, you will realize that
something important is missing from what you said: the reference
signal, or reference condition. Why do you react to the angry
dog? Is it because you have learned to perform certain motions
when certain visual images are projected onto your retina? Or is
it because you don't want the dog to bite you?
When you see an angry dog, you perceive, from memory and
imagination, possible outcomes. Among these outcomes is a dog
biting you. So far that is simply information. But then you
compare that outcome with a desired outcome, a reference signal,
which in this case is set to a very low value, probably zero. You
wish the perception of being bitten by a dog to be zero. So
perceiving the angry dog leads to an imagined perception of being
bitten, and the error signal is turned into actions that would
tend to bring that perception to zero. The actions, such as
hopping into a car and closing the doors and windows, leads now
to a projected situation in which the perception of being bitten
is at an acceptable level, zero. So you have corrected the error.
It wasn't a reaction to a stimulus at all.
In talking about S-R situations it's easy to forget to mention
the reference signal (reference level, reference value, reference
condition -- there are many ways to say it). An animal behaves
when given food only if it is hungry -- has a nonzero reference
level for eating food. Perceptions in themselves have no value;
they are simply reports on the current state of affairs. Their
value comes from the reference signal, which specifies how much
of any particular perception is wanted. If a perception happens
to match a reference value, it calls for no action at all. You
don't demand vanilla ice cream when there is a dish of vanilla
ice cream in front of you. You might demand a spoon.
In most S-R situations, there is an unmentioned reference
condition that explains the reaction to the stimulus.
---------------------
From your reply to Martin:
i.e. for hearing & speech
channel capacity C = B log2(1+S/N)
[C]=1/s
B=channel bandwidth (Hz)
S=transmission power (Watts)
N=noise power (Watts)
How strange. The last time I tuned in to a Finnish short-wave
station, I got a good signal to noise ratio but my channel
capacity was zero.
------------------------
Are not PCTers replacing the S-R behaviorism by PCT
behaviorism. Life is more than behaving.
Life is experiencing, isn't it? That's why we speak of
controlling perceptions, not actions. We use whatever actions are
required to make our experiences be what we want them to be, when
we can.
But there are things we can't explain. The main one is
consciousness. There are other things we might some day explain
with PCT but haven't yet explained, because we know too little
about the details of complex perceptions. A few things we can
explain now, quite well. But we really have little interest in
behavior. Behavior is just output. What human beings are
concerned with is not what they do to the world, but what they
experience as a result of what they do.
-------------------------------
... one of the main functions of the nervous system is to
reduce redundancy from the sensory information. We can make a
artificial system, which collects huge amounts of sensory
information just for controlling one on/off function (nuclear
defense e.g.).
I would say that is one of the main _effects_ of the nervous
system. I don't buy the idea that the nervous system spends all
its time trying to reject all that excess information that's
trying to get in. It takes a great deal of work for the nervous
system to construct a perception that makes any sense; there is
no surplus of perceptions. I think that all perceptions were
constructed for the purpose of making life better for the
organism, even if they're not always being controlled.
When you use huge amounts of information for controlling a single
function, this probably means you don't understand what's going
on, and have very poor control of whatever it is you're trying to
control. And anyway, if huge amounts of information have been
gathered, surely they can be used for more than one purpose.
-------------------------------------------------------------
Best,
Bill P.